The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local developer utilities such as leaks, xcrun simctl, vmmap, and heap via subprocess.run. These operations are essential for memory graph analysis and are implemented using list-based argument structures that prevent shell-injection vulnerabilities. Parameters like the process ID and memory addresses are validated or extracted using strict regex patterns before execution.
[PROMPT_INJECTION]: The skill processes output from Apple's memory analysis tools, which constitutes an indirect prompt injection surface if the application being analyzed contains malicious strings. 1. Ingestion points: Text output from leaks --list, traceTree, and other CLI tools processed in scripts/summarize_memgraph.py. 2. Boundary markers: The skill instructs the agent to treat tool output as data, though no explicit XML-style delimiters are used in the prompt templates. 3. Capability inventory: Local file system read/write and execution of memory inspection tools. 4. Sanitization: The scripts use specific regex patterns (LEAK_LINE, TOTAL_LINE) to parse relevant data and implement a bounded_text reader that limits raw input to 1 MiB, mitigating risks from maliciously oversized or malformed tool outputs.

ios-memgraph-analysis